The key results generated from this investigation are summarized in the bulleted paragraphs below.
?? The Panama Canal watershed contains a strongly spatially structured tree flora with high beta- diversity. Primary floristic differences at the watershed scale appear to be well correlated with established Holdridge Life Zones.
?? Tree composition within Holdridge Life Zones appears to be highly variable and dominated by the trans-isthmian climatic gradient. In some situations, soil attributes can take priority over local precipitation conditions and generate distinctive tree compositions.
?? The climatic gradient is strongly expressed through the composition of individual tree species.
Tree families are relatively well mixed across the isthmus and do not exhibit strong gradient patterns.
?? Floristic nestedness is low along this resource gradient as species turnover very rapidly across the isthmus.
?? Opportunities for increasing relative rates of tree species accumulation within a given geographic area with knowledge of local soil conditions, floristic assemblages, and geographic setting (e.g., distance between reserves).
Watershed floristic patterns
Both agglomerative, hierarchical cluster analysis (Figure 7) and NMDS analysis (Figure 8) suggest that plots in locations around the edges of the Panama Canal watershed have very low coefficients of similarity with those immediately along the line of the Canal. Visual inspection of two- dimensional NMDS scores for the complete set of 54 plots indicates a long cloud of points stretching
from the dry sites at Cocoli to the wet Caribbean plots at Fort Sherman with a more diffuse group containing the outer watershed plots stretching off alo ng a second axis. Cluster analysis (Ward’s Method of agglomeration with Euclidean distance measures for 54 plots, chaining = 2.05) created a dendrogram indicating that the first major subdivision of the data separates plots on the wet and dry ends of the isthmus, and the next division isolates all of the outer watershed plots (specifically m31 through m39). It is noteworthy that the clustering and NMDS results compliment each other as the classification routine considered species frequency (Euclidean distance measures) while the NMDS solution was based on presence-absence (Sorensen similarity coefficients). These preliminary data suggest that plot composition immediately adjacent to the Canal vary along local ecological gradients, while broad floristic associations across the watershed are more associated with bioclimate as indicated by Holdridge Life Zones.
i. Spatial Structure
The Mantel test indicated an association between geographic distance and floristic similarity. Both the asymptotic approximation method (t-value = 8.282, p-value = 0.0000, infinite degrees of freedom) and Monte Carlo simulations (1000 trials, Zobserved >> Zaverage, p = 0.001), indicate that the positive association is highly significant. The Monte Carlo results indicated that the observed Z score was much greater than the expected score generated from an average of the randomized trials. This result supports a strong positive relationship where floristic distance (1- Sorensen similarity) increases directly with increasing geographic separation.
ii. Species accumulation across the watershed
Condit et al. (1996) performed a series of species area calculations for tropical forests, including lowland Panama centered on BCI. They suggest that species area curves should show a
the Canal Area plots (Cocoli to Fort Sherman) indicates a rapid increase followed by a long, slowly rising trend in species accumulation regardless of the order in which the plots are added (Figure 4). If the outer watershed plots are added to the end of the mainline species area curve, again in any order, the slope increases dramatically indicating the accumulation of many new species. The sudden change in slope between these curves is indicative of a demonstrable change is rates of species accumulation, i.e., the outer watershed plots contain many species not found along the Canal. For comparison, the 50-ha plot on BCI contain 303 species, when combined the 45 1-ha Canal area plots contain 417 species, and the addition of only 9 more hectares from the outer watershed plots increases the total to 821 species.
Gradient dynamics along the Panama Canal
i. Ordination of the 45 Canal Area plots
A strong environmental gradient well correlated with climatic indices was identified in the 45 Canal Area plots. This observation was supported by both DCA and NMDS. The application of Principle Components Analysis to center the two-axis NMDS scores provided information on the structure of the data. Table 3 illustrates the dominance of the first axis in explaining NMDS variation among the plots. Results from DCA support the NMDS ordination by indicating a strong, first axis eigenvalue of 0.7 for the 45 plots. The results compliment each other as the DCA provides a measure of species turn -over and axis length; while, the NMDS ordination provides better separation between plots in the ordination space and more potential for identifying secondary contributing factors (e.g., geology).
Table 3. Variance extracted from the first two axes of PCA on NMDS scores for 45 Canal Area monitoring plots (PC-ORD, version 3.18).
Axis Eigenvalue % of Cumulative %
1 36.07 80.15 80.15
2 8.93 19.85 100.00
ii. Environmental correlations with indirect floristic gradients
The centered NMDS axis scores were correlated with the continuous, quantitative environmental variables, including: UTM coordinates x and y, elevation, total precipition, median precipition, and total May precipitation. Table 4 indicates the least-square fit correlation coefficients for the NMDS first axis.
Table 4: Correlations with continuous environmental variables and NMDS axes.
Axis 1 2
Variable R r-
squared
R r-
squared
Utm x -.558 .311 .541 .292
Utm y .557 .310 -.571 .326
Elevation .294 .086 -.131 .017 Total ppt. .606 .367 -.563 .317 Median ppt. .607 .369 -.570 .324
Figure 9 illustrates correlations between first axis NMDS and DCA scores and interpolated precipitation values. These diagrams show least-square best fit lines plotted for each axis. Although the NMDS scores provided enhanced separation of the individual plots relative to the DCA output, relationships to categorical variables were still difficult to identify. The lack of replicates makes case specific patterns (e.g., similarities or differences between pairs of plots) difficult to generalize. The categorical variables for geology, age, and topography are superimposed on the centered NMDS scores in Figure 10. Several instances standout for further investigation, including the limestone L1 plot and the ordination outliers plots m25 and m26. L1 (3100 mm/year ppt.) is surrounded, in
ordination space, by plots characterized by much drier conditions on the Pacific coast (~1800 mm/year
ppt.); while, m25 and m26 appear more similar to plots tens of kilometers away on the Caribbean Coast than those close by.
Multiple regression models were explored in an attempt to explain relationships between environmental variables and ordination axis scores. The most successful models are detailed below, and it appears that the combination of a precipitation index with geology and elevation fit the observed axis scores with a multiple r-squared of 0.4553.
Model 1:
centered NMDS 1 = (interpolated median annual ppt, mm) + (geologic unit)
Coefficients:
Value Std. Error t value Pr(>|t|) (Intercept) -3.5427 0.5535 -6.4009 0.0000 medppt 0.0015 0.0002 7.5882 0.0000 geo -0.0715 0.0239 -2.9970 0.0046
Residual standard error: 0.4228 on 42 degrees of freedom Multiple R-Squared: 0.6863
F-statistic: 45.94 on 2 and 42 degrees of freedom, the p-value is 2.677e-011
Correlation of Coefficients:
(Intercept) medppt medppt -0.9781 geo -0.4899 0.3328
Model 2:
centered NMDS 1 = (interpolated cumulative May ppt, mm) + (geologic unit)
Coefficients:
Value Std. Error t value Pr(>|t|) (Intercept) -2.6203 0.4426 -5.9206 0.0000 newmay 0.0188 0.0025 7.4599 0.0000 geo -0.0702 0.0242 -2.9063 0.0058
Residual standard error: 0.427 on 42 degrees of freedom Multiple R-Squared: 0.6801
F-statistic: 44.64 on 2 and 42 degrees of freedom, the p-value is 4.038e-011
Correlation of Coefficients:
(Intercept) newmay newmay -0.9648 geo -0.5361 0.3411
Model 2 illustrates that replacing the total annual precipitation variable with the interpolated dry season precipitation value did not improve the fit with the first axis of variation. These models suggest that the interpretation of the ordination scores is relatively insensitive to the choice of precipitation index. It is interesting to note that the only other significant predictive variable was substrate geology. Incorp orating stand age in the model provides a little additional information, but its coefficient is not significant (p = 0.1078).
iii. Evaluation of the rainfall gradient across taxonomic scales
DCA ordination of tree species indicates a strong gradient with plots densely clustered along the first axis (Figure 11, Axis 1, eigenvalue: 0.70). Aggregating species to genera produced a weaker gradient with more dispersion of plots into second and higher order axes (Figure 12, Axis 1, eigenvalue: 0.59). Families show the weakest response to the precipitation gradient (Figure 13, Axis 1, eignvalue 0.34). The trend is from a strong, linear group of plots in ordination space to a more diffuse cloud of points with less structure with respect to ordination axes. Overall, the strength of response to the precipitation gradient identified by DCA declines with increasing taxonomic aggregation.
These patterns are interesting, but they can not be fully decoupled from species -level gradient responses. Most genera (165 genera out of a total of 231 genera) and many familes (23 families out of a total of 69 families) contain a single species. Taxonomic aggregation has no impact on these taxa, and species -level responses will continue to contribute to the higher level ordinations.
these monotypic taxa. The most common polytypic genera included: Ficus (17 species), Inga (16 species), Pouteria (12 species), Eugenia (11 species), Miconia (11 species), Ocotea (9 species), Guarea (7 species), Protium (7 species), Brosimum (6 species), Casearia (6 species). The families contributing the greatest number of species included: Fabaceae (41 species), Moraceae (33 species), Sapotaceae (20 species), Rubiaceae (19 species), Lauraceae (18 species), Myrtaceae (18 species), Euphorbiaceae (16 species), Melastomataceae (15 species), Flacourtiaceae (15 species).
It is also noteworthy that the spatial structure identified across the watershed breaks d own with increasing taxonomic aggregation. Mantel tests (as described in Methods: Watershed Floristic Patterns, iii. Tests for spatial structure) were applied to examine spatial structure in only the 45 plots used for gradient and taxonomic analysis (Table 5). The tests indicate that while species and genera - level representations of the plots are relatively well organized across the landscape, this structure broke down when floristic data was aggregated to the level of families.
Level of aggregation
Mantel’s asymptotic approximation
Equivalent t-statistic
p-value
Species 0.098 2.268 0.024
Genus 0.084 1.825 0.068
Family 0.862 0.862 0.389
iv. Diversity and nestedness
The lowland forest was observed to be highly diverse, and the median frequency of occurrence in monitoring plots was 1 for all species (Figure 14, Figure 15). The majority of these species appear to be extremely rare with maximum observed abundances on the order of less than 1 individual in 1,000 stems sampled. Local abundance appeared to be a predictor of regional frequency of occurrence, where locally common species were also more wide-spread than their locally rare counterparts (Figure 16). This pattern may reflect a sampling artifact driven by the relative
probabilities of sampling common versus rare species. Comparing maximum local abundance, as opposed to average local abundance, may somewhat reduce this convolution, but it does not eliminate it.
Few strong correlates with plot diversity were developed during exploratory analysis;
however, species richness appeared to be related to estimated stand age (Figure 17). Old-growth stands were significantly more diverse than secondary and young forest types (p < 0.01, two -side, two-sample t-test for unequal variances, with df = 26).
Monte Carlo simulations indicate that more floristic nesting exists than would be expected by chance alone, but the degree of this nesting is far less than many biological systems (e.g., small mammal populations on isolated mountain tops, Kadmon 1995). When optimally packed by the NestCalc algorithm, the plots on BCI and in the surrounding forest contain the most ubiquitous species and to a very limited degree both Cocoli and Fort Sherman represent somewhat depauperate samples of this flora (Table 5). It is interesting to note that the flora appears somewhat nested from the interior of the isthmus toward the coasts.
Values for species richness and Shannon diversity are presented for each plot in Table 6.
The plots are organized by their nestedness ranking (column 1). Based only on presence-absence data, the plots with high ranks (low numbers in column 1) contain relatively rich samples of wide- spread species. Conversely, plots with low relative ranks (high numbers in column 1) contain a more specialized flora. Diversity, as measured both by richness and Shannon diversity, roughly correlates with the nestedness ranking.
Table 6: The ranking below was produced with NestCalc and illustrates the relative nestedness of the samples from each plot. Plots with low rank numbers contain more complete sets of the tree flora than plots with higher rank numbers. See the main text in the Methods section for a more complete description of the ranking procedure.
Rank Plot Richness (S)
Shannon diversity (H')
Stems (>= 10 cm DBH)
Location
1 By1 93 3.42 526 Interior
2 Bs1 98 3.97 408 Interior
3 M9 107 3.91 503 Interior
4 M14 92 3.93 381 Interior
5 M15 91 3.89 457 Interior
6 Bhp 84 3.53 424 Interior
7 Blp 90 3.76 409 Interior
8 M7 93 3.95 380 Interior
9 M8 94 3.54 560 Interior
10 M16 90 3.70 467 Interior
11 M19 89 3.66 519 Interior
12 Bs2 87 3.74 407 Interior
13 M18 86 3.89 429 Interior
14 L4 94 4.06 450 Interior
15 S1 81 3.75 531 Caribbean
16 M10 78 3.65 403 Interior
17 M12 74 3.33 520 Interior
18 L2 84 3.90 409 Interior
19 M6 78 3.62 480 Interior
20 By2 75 2.65 597 Interior
21 S3 75 3.81 526 Caribbean
22 M20 90 3.70 534 Interior
23 S0 88 3.83 480 Caribbean
24 M22 75 3.37 508 Interior
25 M5 71 3.43 364 Interior
26 M25 84 3.80 593 Interior
27 M11 75 3.43 449 Interior
28 M21 78 3.76 405 Interior
29 L3 74 3.82 365 Interior
30 M17 63 3.02 461 Interior
31 M26 76 3.41 485 Interior
32 S4 70 3.06 954 Caribbean
33 S2 65 3.55 484 Caribbean
34 M23 60 2.70 579 Interior
35 M27 61 3.45 393 Interior
36 M28 62 3.33 408 Interior
37 M29 65 3.33 355 Pacific
38 M13 60 2.44 647 Interior
39 M24 60 2.95 557 Interior
40 M30 64 3.16 302 Pacific
41 L1 63 3.13 400 Caribbean
42 C3 57 3.29 241 Pacific
43 C1 50 3.17 288 Pacific
45 C2 49 2.96 257 Pacific